FISCAL DECENTRALIZATION AND THE QUALITY OF GOVERNMENT: EVIDENCE FROM PANEL DATA (*)

Transcripción

1 FISCAL DECENTRALIZATION AND THE QUALITY OF GOVERNMENT: EVIDENCE FROM PANEL DATA (*) Autores: Andreas P. Kyriacou (1) Universitat de Girona Oriol Roca-Sagalés (2) Universitat Autònoma de Barcelona P. T. N. o 3/08 (*) The authors acknowledge the Instituto de Estudios Fiscales (Ministerio de Economía y Hacienda) and the Robert-Shuman-Center for Advanced Studies (European University Institute) for their financial and academic support respectively. Andreas Kyriacou benefited from project SEJ (Ministerio de Ciencia y Tecnología) while Oriol Roca-Sagalés from projects SEJ (Ministerio de Ciencia y Tecnología) and 2005SGR-117 (Direcció General de Recerca). They also thank Daniel Treisman for sharing his data on decentralization with them. (1) Departament d Economia, Universitat de Girona. Campus de Montilivi, 17071, Girona. (2) Departament d Economia Aplicada, Universitat Autònoma de Barcelona. Edifici B, Bellaterra 08193, Barcelona. N.B.: Las opiniones expresadas en este trabajo son de la exclusiva responsabilidad de los autores, pudiendo no coincidir con las del Instituto de Estudios Fiscales. Desde el año 1998, la colección de Papeles de Trabajo del Instituto de Estudios Fiscales está disponible en versión electrónica, en la dirección: >

2 Edita: Instituto de Estudios Fiscales N.I.P.O.: I.S.S.N.: Depósito Legal: M

3 INDEX 1. INTRODUCTION 2. THEORETICAL PRIORS 3. EMPIRICAL PRIORS 4. DATA AND EMPIRICAL METHODOLOGY 5. EMPIRICAL RESULTS 6. DISCUSSION BIBLIOGRAPHY SÍNTESIS. Principales implicaciones de política económica 3

4

5 ABSTRACT In this paper we focus on the relationship between fiscal decentralization and government quality. In a sample of 29 developing and developed countries over the period , fiscal decentralization has a positive effect on institutional quality but this effect diminishes as countries become wealthier. Moreover, the positive effect of fiscal decentralization on government quality is reduced by electoral and decision-making decentralization in poor and medium income countries whereas these forms of decentralization seem to improve the impact of fiscal decentralization on government quality in rich countries. JEL Classification: H11, H77, K42. Keywords: Quality of government, fiscal decentralization, political decentralization, panel data. 5

6

7 Instituto de Estudios Fiscales 1. INTRODUCTION In this paper, we empirically investigate the relationship between fiscal decentralization and government quality for a sample of twenty-nine poor and rich countries over the period The quality of government is a multidimensional concept (La Porta et al., 1999). In general, government quality improves if the public sector does not distort the proper functioning of the private sector (respects property rights, does not over-regulate), is an efficient administrator (low corruption, less bureaucracy, high tax compliance), provides public goods (education, health, infrastructures, etc.) and, finally, allows for and protects political freedom. A good number of empirical studies have pointed to the importance of good government for economic growth (Knack and Keefer, 1995; Mauro, 1995; Hall and Jones, 1999; Acemoglu et al., 2001; Salinas and Salinas, 2007). But the quality of government has itself been the object of analysis in an effort to understand why some countries have better governments than others. In this context, La Porta et al. (1999) point to the importance of economic, political and cultural factors in explaining international differences in government quality. More recently, Adserà et al. (2003) and Persson et al. (2003) have shown the importance for government quality of periodic elections, a well informed electorate and well designed electoral rules 1. Another factor which may determine variation in government quality is the degree of fiscal decentralization. As is explained in section 2 below, the theoretical literature provides arguments both for and against decentralizing fiscally in an effort to contribute towards better government. Similarly, the lion s share of empirical studies that have examined the effect of fiscal decentralization on government quality have not clearly established the sign and significance of this relationship as is explained in section 3. This study will add to this empirical literature by estimating the impact of fiscal decentralization on government quality based on a panel of countries. It improves on previous studies which are mostly cross-section based not least because it increases the sample size and thus improves the precision of the estimates. Moreover, and as discussed in section 4, the use of fixed effects estimation corrects one serious problem affecting previous cross-section estimates namely, omitted variable bias. The obtained results, discussed in sections 5 and 6, point to a positive relationship between fiscal decentralization and government quality in the case of poor to medium income countries. Another notable finding which emerges 1 See Aixalà and Fabro (2007a) for a cross-section analysis of the determinants of government quality as well as Aixalà and Fabro (2007b) for a review of the literature of the impact of government quality on economic growth. 7

8 from the analysis is the impact of political decentralization on this relationship. Political decentralization seems to erode the positive effect of fiscal decentralization on government quality in poor countries while it improves its impact on the quality of government in rich countries. 2. THEORETICAL PRIORS There are several theoretical arguments why decentralization may improve government quality. First, decentralized government will be better informed about local conditions and so can better satisfy citizen preferences (Oates, 1972). Second, citizens themselves may be better informed about local government activity and, assuming that local politicians are locally elected rather than centrally appointed, good (or bad) performance in such a setting can be directly rewarded (or punished) (Seabright, 1996; Tabellini, 2000). Third, in a fiscally decentralized setting with interjurisdictional mobility and thus competition, we would expect locally elected governments to offer public goods more efficiently (responsive to local demands and at lower cost) and also to be less corrupt since economic agents would flee more corrupt regions (Hirschman, 1970; Brennan and Buchanan, 1980; Weingast, 1995; Montinola et al., 1995). The capacity of economic agents to choose among regulatory regimes in a competitive setting will tend to drive down the degree of corruption whereas corruption is likely to increase in a setting characterized by numerous public monopolies all of which must be satisfied to undertake any particular economic activity (Shleifer and Vishny, 1993). In a competitive setting, voters can use the performance of each jurisdiction as a benchmark and this should increase efficiency in public good supply (Salmon, 1987; Breton, 1996). The inter-jurisdictional competition literature emphasizes the importance of tying local revenues to expenditures for the proper functioning of competition since vertical transfers may create incentives for local officials to ignore competitive pressures for better management (Oates, 1999; Fisman and Gatti, 2002b). Similarly, in the context of revenue sharing schemes, assigning local government a greater share of revenue from locally-originating taxes may lead them to limit costly regulations, corruption as well as provide public goods more efficiently since these reduce economic activity and thus eventually reduce its tax base (Zhuravksaya, 2000; Jin et al., 2005). More generally, sub-central governments will tend to be more inefficient if they face soft rather than hard budget constraints, or, in other words, if they can count on the central government for future bail-outs (Qian and Rolands, 1998). On the other hand, several arguments have been made explaining why fiscal decentralization may deteriorate government quality. For example, if capital is 8

9 Instituto de Estudios Fiscales too mobile, local governments may not be able to collect sufficient taxes to provide basic public goods (Keen and Marchand, 1997; Oates, 1999). Further, locally elected governments may indulge in rent-seeking behaviour if more immobile factors such as farmers, workers or consumers are more numerous than mobile ones such as investors (Rodden and Rose-Ackerman, 1997). Moreover, the benign effects of competition depend on the assumption that sub-central jurisdictions are homogeneous with regard to their factor productivity. If they are heterogeneous, competition may mean the outflow of capital from less endowed regions leading governments in these regions to give up on business-friendly policies and to dedicate themselves to predation instead (Cai and Treisman, 2005). In addition, local governments may also compete for capital by helping agents cheat the central government in ways that reduce the latter s capacity to enforce regulations and collect taxes (Cai and Treisman, 2004). Also, although assigning local governments a greater share of revenue may improve their incentives, it may at the same time weaken the incentives of central government by reducing its stake in economic development since its share of locally generated revenues is reduced (Treisman, 2006). Additionally, more decentralized polities may involve duplication and thus a waste of resources (Rousseau, 1762). Moreover, local officials are more susceptible to corruption than central government ones because of intimacy and frequency of interaction and greater discretion on the part of local governments (Tanzi, 1995; Prud'homme, 1995), because national office being more prestigious and powerful it will be more intensely monitored (Prud'homme, 1995; Tabellini, 2000) or because local officials are more susceptible to capture by local economic interests partly because voters are more uninformed due to lower media attention and because interest groups find it easier to overcome freerider problems (Bardhan and Mookherjee, 2000). In this vein, Blanchard and Shleifer (2001) suggest then need for a degree of political centralization such as for example the right of central government to appoint or fire regional governors or the creation of national parties exercising influence on regionally elected governors so as to deal with the possibility of capture. Alternatively, the combination of strong national parties and elected rather than appointed sub-central government politicians makes them sensitive to national and local preferences respectively and is thereby likely to improve the effectiveness of fiscal decentralization (Riker, 1964; Enilolopov and Zhuravskaya, 2007). 3. EMPIRICAL PRIORS Most empirical studies on the impact of decentralization have focused on one aspect of government quality namely, corruption. Thus, De Mello and Barenstein (2001) control for GDP per capita and population and find that more fiscally 9

10 decentralized countries tend to be less corrupt. Fisman and Gatti (2002a) find a similar relationship but controlling for more factors (GDP per capita, civil liberties, population and government share of expenditure of GDP). They moreover find that when fiscal decentralization is considered then whether a country is federal or not has no effect on corruption. Treisman (2002a) finds that if one includes a control for the percentage of Protestants in the population in Fisman and Gatti s estimations, the negative relationship between fiscal decentralization and corruption disappears - he finds no statistically significant relationship between fiscal decentralization and corruption. All these empirical studies calibrate the impact of fiscal decentralization on corruption based on cross-section analyses 2. Several empirical studies have gone beyond corruption to consider other aspects of government quality in order to investigate the importance of fiscal decentralization. Estache and Sinha (1995) calibrate the effect of expenditure decentralization on infrastructure provision in a panel of both developing and developed countries. They find that decentralization is related to more infrastructure spending in sub-central levels in developing countries and, to a lesser extent, in developed ones. Robalino et al. (2001) also use panel data and find that more fiscally decentralized countries (again expenditure decentralization) have lower mortality rates. In another study, (Khaleghian, 2003) shows that fiscal decentralization (measured in terms of both a binary variable and as expenditure and revenues in the hands of sub-central governments) is related to higher immunization coverage in a panel of low and middle-income countries. In a cross-section study, Treisman (2002a) finds expenditure decentralization ambiguously related to public good provision (positively with infrastructure and negatively with health and education). Defining government quality more widely, Huther and Shah (1998) find that fiscal decentralization is associated with enhanced quality of government as measured by citizen participation, political and democratic accountability, social justice, improved economic management and reduced corruption. However, their study suffers from omitted variable bias since they only provide simple correlations among these variables. Finally, based on both cross-section and panel data of low and high income countries, Dreher (2006) estimates the relationship between expenditure and revenue decentralization and quality of government measured as respect of law and order, the cost of opening up a new business and judicial independence while controlling for GDP, population and civil liberties. He finds that law and order is more likely with a higher share of sub-national revenues for the full 2 For a survey of the causes and consequences of corruption see Lambsdorff (2006). For a review of some of the theoretical and empirical literature on the relation between decentralization and corruption see Bardhan and Mookherjee (2006). 10

11 Instituto de Estudios Fiscales sample and low income countries but not for high income countries. The same happens with the cost of starting a business. Judicial independence increases with sub-national revenue or expenditure share for both high income countries and the full sample. In this paper we estimate the impact of fiscal decentralization on government quality more widely defined based on panel data. As will be more fully explained in the following section, our definition of government quality is wider than that used by Dreher (2006) since it goes beyond law and order to include corruption and bureaucratic quality. We reduce the risk of possible omitted variable bias present in previous studies by including more time-variant control variables as suggested by the literature as well as several interaction terms. In addition, we correct our estimates both for serial autocorrelation and cross-section heteroscedasticity to avoid overestimating the significance of our explanatory variables. 4. DATA AND EMPIRICAL METHODOLOGY In this section we review how both fiscal decentralization and quality of government have been measured in empirical work with a view to identifying the strengths and weakness of the available data. We moreover introduce the empirical methodology which is employed in the following section and review the set of control variables which have been used in related empirical literature examining the relationship between fiscal decentralization and government quality. The indicator of government quality employed in this paper comes from the International Country Risk Guide (ICRG) as developed by the Political Risk Services Group since the early 1980s in an effort to assess the political, economic and financial risks in more than 120 countries. The index is based on the analysis of a worldwide network of experts, and is subject to a peer review process to ensure the coherence and comparability across countries. Different variables from the ICRG data have been used to measure the determinants of government quality since La Porta s et al. (1999) seminal piece on the subject. We focus on three dimensions from this source namely corruption, rule of law and bureaucratic quality. Corruption refers to the demand for bribes by political and administrative bodies as well as patronage, nepotism, job reservation, favors-for-favors, etc. The variable law and order assesses the strength and impartiality of the legal system as well as popular observance of the law. Bureaucratic quality refers to the strength, expertise, autonomy and recruitment and training mechanism of the civil service. The quality of government indicator used here is the average of these three dimensions and is 11

12 scaled from 0 to 1 where a higher number implies higher government quality. Averaging the dimensions has the advantage of reducing the influence of idiosyncratic ratings stemming from any one dimension. On the negative side however, it does not allow us to separate the effect of each dimension. Fiscal decentralization refers to the distribution of tax revenues and expenditures among the different tiers of government. A system is more fiscally decentralized the greater the proportion of tax-revenues and expenditures owned by lower tiers of government. The most widely used measure of fiscal decentralization in the literature is provided by the IMF s Government Finance Statistics (GFS) and refers to expenditure or revenue share of sub-central governments in total public expenditures or revenues (Expshare and Revshare) or as a proportion of GDP (ExpGDP and RevGDP). But these general definitions are problematic in that they do not reflect on whether sub-central governments own these resources (Ebel and Yilmaz, 2002; Rodden, 2004; Martinez-Vazquez and McNab, 2006). Thus, expenditures that are mandated by the central government appear as sub-national expenditures in the GFS while revenues received from grants by the central government appear as sub-central revenues regardless of whether they are conditional or unconditional. As a result, expenditure and revenue accruing to sub-central levels of government will tend to overestimate the degree of fiscal decentralization (World Bank, 2004) 3. Despite their shortcoming, these indicators are available for the widest coverage of countries and time periods and it is for this reason that we too employ them in this paper. They give rise to a balance panel of 29 countries for the period 1984 to 1997 (see appendix 1 for the list of countries). The GFS also contain data on other fiscal decentralization variables but with an even more limited cross-section and temporal availability. One line of data refers to vertical imbalance (VIM) or the degree to which sub-national governments rely on central government transfers to support their expenditures (measured as intergovernmental transfers as a share of subnational expenditures). Another calculates transfers share as transfers to subcentral governments as a share of total sub-national revenues and grants (Trashare). However, these measures do not distinguish what proportion of transfers is conditional versus general purpose. The GFS also contain information about sub-national tax and non-tax revenues. For example, tax share measures sub-national tax revenue as a percentage of total sub-national revenues (Taxshare). But it does not distinguish whether they are collected through shared taxes, piggybacked taxes, and locally determined "own-source" 3 Rodden (2004) points out that in the GFS data Denmark comes out the 3rd most decentralized country in the world (even more decentralized than the US) though the central government regulates almost every aspect of local government finance. The same goes with Nigeria (7th position) although the states were really administrative outposts of the central government. 12

13 Instituto de Estudios Fiscales taxes. Similarly, no consideration is given to whether the central government has the power to set the rate and the base, leaving sub-national governments just to collect revenues labeled as own-source (Rodden, 2004). The OECD (1999) tackles these data problems for OECD countries generating an analytical classification of sub-central taxes according to the degree of discretion of subcentral government. Based on this classification, Stegarescu (2006) generates several indicators of decentralization across 23 OECD countries for the period including a measure of tax decentralization which refers to subcentral government tax revenue on which the sub-central government determines the tax rate and/or base as a proportion of total government tax revenue taxes (Taxdec) 4. The just mentioned variables are available for the 15 high income countries from our total sample of 29 countries. As can be appreciated in the correlation matrix in table 1 below there are four variables which are closely correlated namely, ExpGDP, Expshare, RevGDP and Revshare. Stegarescu s Taxdec measure is also closely correlated to the IMF statistics on the revenue side but less so on the expenditure side as we would expect. Table 1 CORRELATION MATRIX FOR THE FIFTEEN HIGH INCOME COUNTRIES. DECENTRALIZATION VARIABLES ExpGDP ExpShare RevGDP Revshare Taxdec Taxshare Trashare VIM ExpGDP Expshare RevGDP Revshare Taxdec Taxshare Trashare VIM All correlations are statistically significant at the 1% level. Another highly correlated pair of fiscal decentralization indicators are VIM and Trashare which is not surprising since they both capture sub-central 4 Even with such an indicator, Rodden (2004) alerts us that ideally one should take into account the possibility that central governments may restrict fiscal autonomy through formal limitations on borrowing (not only through conditional grants and regulation of tax rate and base). 13

14 government dependence on central transfers on the expenditure and revenue sides respectively. Both these variables are moreover highly negatively correlated with Taxshare which again is what we would expect since it is the negative image of these two variables it captures tax-autonomy whereas they capture the tax-dependence of the sub-central governments. These three variables are less correlated with Taxdec as well as Revshare and RevGDP and much less correlated with the general expenditure variables offered by the IMF. Our empirical strategy is based on fixed effects estimation for panel data. Bringing fixed effects estimation to bear on the relationship between fiscal decentralization and government quality has one important advantage over previous cross-section estimations exploring this relationship namely, the ability to control for cross-country heterogeneity due to time constant (or slowly changing) observable or unobserved factors (cross-section fixed effects) as well as the ability to control for period-specific factors common to all cross-section units (period-fixed effects). Time constant factors include many variables which have been typically employed in cross-section analyses of the determinants of quality of government such as ethno-linguistic fractionalization, legal origin, religion, latitude, country area and continental or regional dummies. Given the possibility that these variables, and other non-observable ones, may be related to both fiscal decentralization and government quality, their exclusion would induce omitted variable bias, affecting both the estimated impact of decentralization and its statistical significance. Bardhan and Mookherjee (2006) identify this form of bias as one of the important methodological problems with existing cross-section studies estimating the effect of decentralization on governance. We lag the RHS variables with their values lagged one period to mitigate problems stemming from possible reverse causality in the panel regressions (see also, Enilolopov and Zhuravskaya, 2007). To allow for arbitrary period serial correlation and period heteroskedasticity between the residuals for a given cross-section, we report Period SUR panel corrected standard errors. Moreover, in order to fathom whether the first order autocorrelation we detect seriously alters the statistical significance of our findings, we also undertake regressions where we correct for this by way of taking four year averages 5. We allow for cross- section heteroskedasticity in these autocorrelation-free regressions by reporting White cross-section corrected standard errors. 5 Davoodi and Zoo (1998) also take averages when measuring the impact of fiscal decentralization on economic growth based on a panel of data. They argue that this is reasonable since they would not expect fiscal decentralization to affect year to year changes in economic growth. This argument could probably be made more strongly for the impact of fiscal decentralization on government quality. 14

15 Instituto de Estudios Fiscales Our fixed effect estimation then reduces the likelihood of omitted variable bias since it eliminates the possible influence coming from both observable and unobservable time constant variables. What fixed effects estimation does not control for are time variant omitted variables affecting government quality and fiscal decentralization. As a result and in tune with the research in this area using both cross-section and panel data, we include the following time variant control variables (available in our sample of countries) in our regressions: real GDP per capita, democracy, size of government, population, and checks and balances (see appendix 2 for descriptions and sources of all the variables). We therefore estimate the following equation: Government Quality it = αi + Yt + β1 Fiscal Decentralizationit 1 + β2xit 1 where i refers to countries and t to years, α i and Y t are country and period fixed effects respectively, X it is the vector of control variables and ε it is the error term. Richer countries tend to be more decentralized (Panizza, 1999; World Bank, 2004) and have better government quality since economic development makes better quality institutions more affordable (Islam and Montenegro, 2002) and will tend to create a demand for better government (La Porta, et al. 1999) perhaps because of income's positive effect on education, literacy and depersonalized relationships (Treisman, 2000). Democracy is positively correlated with decentralization (Panizza, 1999; Treisman, 2002b) and is likely to be negatively related to corruption since it makes politicians more accountable (Adserà et al., 2003). Treisman (2002a) takes population to be an indicator for country size and expects larger (more populous) countries to be more decentralized but also to have lower government quality small states are easier to govern. Mello and Barenstein (2001) suggest that more populous countries may be more heterogeneous. This being the case, we would expect population (as a proxy of heterogeneity) to be positively related to decentralization (Panizza, 1999) and negatively related to quality of government (La Porta et al., 1999). More decentralized countries are likely to have more checks and balances. Checks and balances in the political decision-making process could stabilize good institutions but by locking in the status quo they may just as much impede reforms, including those which may improve the quality of government (Treisman, 2002a). All other things equal, one would expect a larger public sector to be relatively more decentralized. Moreover, Tanzi (1998) expects a bigger state sector to imply greater corruption due to the greater possibility for rents. On the other hand, Fisman and Gatti (2002a) expect a larger government to enjoy economies of scale in the fight against corruption. Similarly, the effect of a larger 15 + ε it

16 state sector on the rule of law and bureaucratic quality is arguably positive insofar as more resources improve the capacity of government to provide such services. In relation to this, higher civil service pay has also been found to reduce bureaucratic corruption (Van Rijckeghem and Weder, 2001). Consistent with previous literature then, we expect a positive effect on government quality coming from income and democracy, a negative effect coming from population while that of government size and checks and balances is more ambiguous. 5. EMPIRICAL RESULTS The variables included in the regressions reported in this section are those which are mostly statistically significant or the exclusion of which seriously changes the results. This means that several variables previously discussed namely, democracy, checks and balances and population are not included because they are not significant and their absence has no bearing on the results. This is despite using several indicators of democracy available (see appendix 2). Especially in the case of democracy and checks and balances, their irrelevance may be due to the limited variation through time for the sample of countries that we examine here. In the tables that follow, the size of government is positively related with the quality of government and this relationship tends to be statistically significant for the sample of high income countries. Tables 2 and 3 report the results for the whole sample of countries measuring fiscal decentralization in terms of expenditure and revenue shares of sub-central governments in total government expenditure and revenues respectively. On the expenditure side fiscal decentralization has a positive and statistically significant effect on government quality (regression 1). The direction and statistical significance of this relationship is reinforced in the regression using the four year averages of the variable as a way of correcting for a first-order autocorrelation process detected (regression 2). Moreover, statistical significance is improved with the inclusion of an interaction variable capturing political institutions as suggested by Enilolopov and Zhuravskaya (2007). It turns out that if regional governments or legislatures are elected rather than appointed (regression 3) or if the constitution grants sub-national regions residual powers or autonomy in certain areas (regression 4), then the positive 16

17 Instituto de Estudios Fiscales impact of fiscal decentralization on government quality would be neutralized for the complete sample 6. Table 2 EXPENDITURE SHARE: WHOLE SAMPLE (1) (2) Means (3) (4) (5) (6) Means (7) (8) Decentralization (0.15)* GDP per capita Size Government GDP per capita* Decentralization Elections* Decentralization Autonomy* Decentralization (0,32) 0,160 (0,24) (0.18)***(0.19)***(0.16)*** (1.29)**(0.49)***(1.17)***(1.19)*** (0,23) 0,103 (0,09) (0,30) 0,161 (0,22) (0,28)*** (0,29) 0,116 (0,22) (0,29)** (0,48)**(0,36)*** 0,245 (0,22) 0,141 (0,08)* (0,45)* 0,236 (0,21) (0,15)**(0,05)*** (0,14)** (0,26)** (0,47)* 0,199 (0,21) (0,14)* (0,28)** Adjusted R Observations Here as elsewhere in this paper, standard errors in parentheses. *, **, *** measures statistical significance at the 10, 5 and 1% levels respectively. All continuous variables are in logarithms. Following Robalino et al. (2001) and Khalegian (2003), we introduce an interaction term between fiscal decentralization and income so as to capture possible differences in performance between rich and poor countries. As a result, we find that the positive effect of fiscal decentralization on government quality is true for poor to medium income countries but not for high income countries as can be seen in regressions 5 to 8. The interaction term is negative and statistically significant and yields a turning point for the base regression (5) of $13,340. The negative impact of political decentralization on the relationship between fiscal decentralization and government quality also appears. 6 Following Treisman (2002a) this refers to electoral decentralization and decision-making decentralization respectively. 17

18 (1) Decentralization (0.12) GDP per capita Size Government GDP per capita* Decentralization Autonomy* Decentralization Elections* Decentralization (0,33) 0,172 (0,25) Table 3 REVENUE SHARE: WHOLE SAMPLE (2) Means (0.24)** (0,23) 0,049 (0,12) (3) (4) (5) (0.15) (0,33) 0,168 (0,24) (0,26) (0.18)** (0,31) 0,167 (0,23) (0,24)** (6) Means (7) (8) (1.01)***(0.79)***(1.00)***(0.99)*** (0,35)** 0,270 (0,21) (0,34)** 0,143 (0,09) (0,36)** 0,266 (0,21) (0,35)* 0,259 (0,21) (0,12)***(0,07)***(0,12)***(0,11)*** (0,22) (0,20) Adjusted R Observations These results are confirmed when using revenue share as our measure of fiscal decentralization (table 3) although their statistical significance is reduced with this measure in those regressions which exclude the interaction between fiscal decentralization and GDP per capita. The turning point of the effect of GDP on the relationship between fiscal decentralization and government quality is now $9,754 for the base regression (5). Again, political decentralization seems to have a negative effect on the relationship between government quality and fiscal decentralization. Since we detect that the impact of fiscal decentralization on government quality differs between richer and poorer countries, we split the sample between low to medium income and high income countries and repeat our estimations (see also Aixalà and Fabro 2007a). Table 4 reports the results for a group of 14 low and medium income countries. We confirm the positive relationship between fiscal decentralization and the quality of government, especially on the expenditure side. However, the negative effect of electoral decentralization or decision-making decentralization (regional autonomy) on this 18

19 Instituto de Estudios Fiscales relationship is also observed. Indeed, in the presence of political decentralization the positive relationship between fiscal decentralization and government quality becomes a negative one. Thus, the result whereby fiscal decentralization positively affects government quality in poorer countries can only be affirmed in the absence of political decentralization. Table 4 LOW AND MEDIUM INCOME COUNTRIES Expenditure Decentralization Revenue Decentralization Decentralization (0.22) GDP per capita (0,49) Size Government 0,231 (0,35) Elections * Decentralization Autonomy * Decentralization (1) (2) (3) (4) (5) (6) (0.25)*** (0,42) 0,222 (0,31) (0,35)** (0.21)*** (0,43) 0,166 (0,31) (0,40)** 0,119 (0,15) (0.49) (0,35) (0.24)** (0,44) 0,218 (0,32) (0,30)*** (0.18) (0,47) 0,233 (0,34) (0,34) Adjusted R Observations Turning now to the group of 15 high income countries (table 5), the positive relationship between government quality and fiscal decentralization disappears. In fact, the direction of this relationship is reversed: more fiscal decentralization reduces government quality in this sample. Interestingly, this negative impact is again altered in the presence of electoral or decision-making decentralization. In rich countries, electoral decentralization mitigates the negative impact of fiscal decentralization while decision-making decentralization may even make this impact a positive one. But all these statements must be immediately qualified since our decentralization variables are neither statistically significant nor are they economically significant compared to their previous values (their impact is reduced by a factor of 10). It seems therefore that the best we can do with the available data is to state that fiscal decentralization does not seem to have a significant effect on government quality in high income countries. 19

20 Table 5 HIGH INCOME COUNTRIES Expenditure Decentralization Revenue Decentralization (1) (2) (3) (4) (5) (6) Decentralization (0.09) (0.13) (0.11) -0,030 (0,04) (0.08) (0.04) GDP per capita (0,14)** (0,14)*** (0,13)*** (0.12)** (0,12)*** (0,12)*** Size Government 0,172 (0,12) 0,153 (0,12) 0,132 (0,12) (0,11)* 0,202 (0,11)* 0,179 (0,11)* Elections * Decentralization (0,17) (0,10) Autonomy * Decentralization (0,16) (0,16) Adjusted R Observations In the case of the rich countries it is possible to check the robustness of the obtained results using the different measures of decentralization available. In table 6, we regress the government quality measure we use on GDP per capita and the size of government while varying the fiscal decentralization indicator. We start with the wider IMF measure (but now spending or expenditure as a percentage of GDP to complement previous results), going through the IMF s measure of tax autonomy followed by its measures of transfer dependence and ending with the OECD measure of tax decentralization taken from Stegarescu (2006) which arguably goes some way towards correcting the over-estimation of decentralization attributed to the IMF measures. (1) ExpGDP Decentralization (0.06) GDP per capita (0,13)*** Size Government 0,142 (0,14) Table 6 DECENTRALIZATION IN HIGH INCOME COUNTRIES GFS Decentralization measures (2) RevGDP 20 (4) Taxshare (5) Trashare (6) VIM OECD (3) Taxdec -0,033 (0,04) (0.02)* (0.04) (0.03) (0.12)*** (0,12)** (0,12)*** (0,12)*** ,208 0,170 0,173 (0,11)** (0,10)** (0,11) (0,11) Adjusted R Observations (0.02) (0,12)** 0,191 (0,11)*